At GTC 2022, NVIDIA introduced the Grace CPU Superchip, a total of 144 Arm v9 architecture CPU cores, connected via NVLink-C2C, providing 900GB/s of connection bandwidth. Seeing that NVIDIA is about to complete the three-chip strategy of CPU, GPU and DPU, Intel also has a bottom card, that is, the first billion billion secondary computing GPU - Ponte Vecchio.
Since the release of the Xe GPU architecture, Intel's ambitions to return to the graphics market are clear. From the entry-level core display to the top acceleration calculation card, it is bound to take it all down. Among them, the first product of Xe HPC high-performance computing architecture is code-named Ponte Vecchio, the first batch of 10 billion subcomputer "Aurora" for the US Department of Energy.
It is reported that Ponte Vecchio uses Intel7, TSMC N7/N5 and other five different processes to manufacture, the number of transistors exceeds 100 billion, the integration of up to 63 Tile unit modules (47 functional), the total package area of 4844 square millimeters, the overall power consumption of up to 600W.
Although no more information has been revealed about the Ponte Vecchio, which positions a high-performance architecture, the relevant models and specifications of the three major graphics card manufacturers are compared. Ponte Vecchio is not expected to be released until 2024, and by that time, NVIDIA will have a B100 based on the next-generation Blackwell architecture, and AMD will have a CNDA3 architecture Oftanbul MI300 series, which is more competitive.
For the current chip giants Intel, AMD and NVIDIA, none of them dare to stop the pace of innovation and development. Intel is vigorously regaining the discrete graphics business; AMD can be used to assemble the dual A platform; NVIDIA in addition to developing a CPU + GPU fusion Grace CPU, but also offload some low-level tasks from the CPU to gpu computing to improve performance. Of course, there is also the menacing Apple self-developed M1 series of chips, with Apple's own software and hardware ecology, debut is the peak. Therefore, the future of high-performance computing has a broad room for improvement, and we look forward to new technologies to amaze every nerve in us.